Advanced Multispectral CT Algorithms
The goal of DECT methods is to estimate a small number of material-specific parameters at each image location, and subsequently use them for material discrimination. A pair of commonly used parameters is the photoelectric and Compton coefficients, which are derived from a physics-based X-ray attenuation model. Several DECT techniques have been suggested since the 1970s [1, 2, 3]. They are mostly targeted at medical applications and do not deal with image artifact mitigation. In the security application, many different materials may be scanned in various degrees of clutter and metal objects are common. In this application, image noise and metal artifacts are more severe and can lead to less reliable estimates of the photoelectric and Compton coefficients.
This project develops a new structure-preserving dual-energy inversion method (SPDE) for the formation of enhanced photoelectric and Compton coefficient images. We form the images as the solution of an optimization problem, which explicitly models the physical tomographic projection process. Metal induced streaking is reduced by appropriately down-weighting unreliable data. A boundary-preserving prior based on  is incorporated to improve object localization. In particular, we estimate a mutual boundary-field along with the photoelectric and Compton images. The boundary field provides accurate object localization and allows smoothing inside the objects. By using the SPDE framework, both noise and metal artifacts in photoelectric and Compton images are greatly reduced. This artifact reduction can lead to more accurate material identification.